Mirrors/yuzu
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

astc_decoder: Optimize the use EncodingData

Browse source 
This buffer was a list of EncodingData structures sorted by their bit length, with some duplication from the cpu decoder implementation.
We can take advantage of its sorted property to optimize its usage in the shader.

Thanks to wwylele for the optimization idea.
This commit is contained in:
ameerj 2021-07-04 22:48:41 -04:00
parent 15c0c213b1
commit 5665d05547
6 changed files with 108 additions and 138 deletions
Download patch file Download diff file Expand all files Collapse all files

50
src/video_core/host_shaders/astc_decoder.comp
View file

@ -10,18 +10,16 @@
#define END_PUSH_CONSTANTS };
#define UNIFORM(n)
#define BINDING_INPUT_BUFFER 0
#define BINDING_ENC_BUFFER 1
#define BINDING_SWIZZLE_BUFFER 2
#define BINDING_OUTPUT_IMAGE 3
#define BINDING_SWIZZLE_BUFFER 1
#define BINDING_OUTPUT_IMAGE 2
#else // ^^^ Vulkan ^^^ // vvv OpenGL vvv
#define BEGIN_PUSH_CONSTANTS
#define END_PUSH_CONSTANTS
#define UNIFORM(n) layout(location = n) uniform
#define BINDING_SWIZZLE_BUFFER 0
#define BINDING_INPUT_BUFFER 1
#define BINDING_ENC_BUFFER 2
#define BINDING_INPUT_BUFFER 0
#define BINDING_SWIZZLE_BUFFER 1
#define BINDING_OUTPUT_IMAGE 0
#endif
@ -64,11 +62,6 @@ layout(binding = BINDING_INPUT_BUFFER, std430) readonly buffer InputBufferU32 {
uint astc_data[];
};
// ASTC Encodings data
layout(binding = BINDING_ENC_BUFFER, std430) readonly buffer EncodingsValues {
EncodingData encoding_values[];
};
layout(binding = BINDING_OUTPUT_IMAGE, rgba8) uniform writeonly image2DArray dest_image;
const uint GOB_SIZE_X = 64;
@ -94,6 +87,19 @@ const int JUST_BITS = 0;
const int QUINT = 1;
const int TRIT = 2;
// ASTC Encodings data, sorted in ascending order based on their BitLength value
// (see GetBitLength() function)
EncodingData encoding_values[22] = EncodingData[](
EncodingData(JUST_BITS, 0, 0, 0), EncodingData(JUST_BITS, 1, 0, 0), EncodingData(TRIT, 0, 0, 0),
EncodingData(JUST_BITS, 2, 0, 0), EncodingData(QUINT, 0, 0, 0), EncodingData(TRIT, 1, 0, 0),
EncodingData(JUST_BITS, 3, 0, 0), EncodingData(QUINT, 1, 0, 0), EncodingData(TRIT, 2, 0, 0),
EncodingData(JUST_BITS, 4, 0, 0), EncodingData(QUINT, 2, 0, 0), EncodingData(TRIT, 3, 0, 0),
EncodingData(JUST_BITS, 5, 0, 0), EncodingData(QUINT, 3, 0, 0), EncodingData(TRIT, 4, 0, 0),
EncodingData(JUST_BITS, 6, 0, 0), EncodingData(QUINT, 4, 0, 0), EncodingData(TRIT, 5, 0, 0),
EncodingData(JUST_BITS, 7, 0, 0), EncodingData(QUINT, 5, 0, 0), EncodingData(TRIT, 6, 0, 0),
EncodingData(JUST_BITS, 8, 0, 0)
);
// The following constants are expanded variants of the Replicate()
// function calls corresponding to the following arguments:
// value: index into the generated table
@ -596,22 +602,16 @@ void DecodeColorValues(uvec4 modes, uint num_partitions, uint color_data_bits) {
for (uint i = 0; i < num_partitions; i++) {
num_values += ((modes[i] >> 2) + 1) << 1;
}
int range = 256;
while (--range > 0) {
EncodingData val = encoding_values[range];
// Find the largest encoding that's within color_data_bits
// TODO(ameerj): profile with binary search
int range = 0;
while (++range < encoding_values.length()) {
uint bit_length = GetBitLength(num_values, range);
if (bit_length <= color_data_bits) {
while (--range > 0) {
EncodingData newval = encoding_values[range];
if (newval.encoding != val.encoding && newval.num_bits != val.num_bits) {
break;
}
}
++range;
if (bit_length > color_data_bits) {
break;
}
}
DecodeIntegerSequence(range, num_values);
DecodeIntegerSequence(range - 1, num_values);
uint out_index = 0;
for (int itr = 0; itr < result_index; ++itr) {
if (out_index >= num_values) {
@ -1110,10 +1110,10 @@ TexelWeightParams DecodeBlockInfo(uint block_index) {
}
weight_index -= 2;
if ((mode_layout != 9) && ((mode & 0x200) != 0)) {
const int max_weights[6] = int[6](9, 11, 15, 19, 23, 31);
const int max_weights[6] = int[6](7, 8, 9, 10, 11, 12);
params.max_weight = max_weights[weight_index];
} else {
const int max_weights[6] = int[6](1, 2, 3, 4, 5, 7);
const int max_weights[6] = int[6](1, 2, 3, 4, 5, 6);
params.max_weight = max_weights[weight_index];
}
return params;